The Agent Skills Directory

[PROMPT_INJECTION]: The skill does not contain instructions that attempt to override agent behavior, bypass safety filters, or extract system prompts. The instructions are focused on architectural auditing tasks.
[DATA_EXFILTRATION]: No network operations (e.g., curl, wget, fetch) or external endpoints are defined. The skill restricts its activity to the local codebase and does not access sensitive system paths like SSH keys or environment secrets.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download or execution of remote scripts, nor does the skill install third-party packages.
[COMMAND_EXECUTION]: The skill uses standard file analysis tools (glob, grep) for the intended purpose of auditing source code. It does not use administrative commands (sudo) or perform persistence-related actions.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface as it reads untrusted user-provided .swift files. While these files could contain malicious comments intended to influence the audit results, the skill lacks the capabilities (such as network access or file writing) to facilitate a meaningful attack.
Ingestion points: Processes local source files via **/*.swift glob patterns.
Boundary markers: Not explicitly defined for the ingested data.
Capability inventory: Limited to reading files and generating a report; no network or write capabilities.
Sanitization: Not applicable to this read-only reporting flow.
[OBFUSCATION]: The skill contains no encoded strings, hidden characters, or homoglyphs. All logic is presented in clear, human-readable text.

axiom-audit-swiftui-architecture